Are Custom Linux Kernels Faster than Stock?
Вставка
- Опубліковано 3 сер 2024
- Are Custom Linux Kernels Faster than Stock?
Benchmarks are done and will be compared using phoronix test suite. We will be analyzing 3 kernels, Liquorix, Mainline, and Xanmod.
Timestamps:
00:00 Introduction
01:34 Xanmod Overview
04:17 Benchmarking Setup
07:15 Benchmarking Run
08:47 Benchmark Results
10:24 Conclusion
---
Xanmod Site: xanmod.org/
Liquorix Site: liquorix.net/
Results: openbenchmarking.org/result/2... .
►► Digital Downloads ➜ www.cttstore.com
►► Reddit ➜ / christitustech
►► Titus Tech Talk ➜ / titustechtalk
►► Twitch ➜ / christitustech - Наука та технологія
Chris, for the next time that you benchmark a kernel for a game, I recommend you to use lower graphics settings. It will show bigger differences but more importantly, it is easily possible that with the highest settings you get 0 difference while with lower settings and possibly a lower resolution you get relevant difference. You want the bottleneck to be at the CPU, not at the graphics card. Trust me on this, I understand hardware well. (For people who interpret this wrong, I do not claim that a CPU-review should necessarily be done with lower settings, that is a different matter for which more factors are relevant.)
Good point, I might try an older game as well.
@@ChrisTitusTech Yeah make sure you have like 120 fps or more i nthe game, then the differences between CPU bottleneck starts showing up.
If you go deep enough down the rabbit hole, you'll soon be rambling about memory bottlenecks and the importance of memory overclocking.
@@ChrisTitusTech and definately add the zen kernel to the list :D
@@BruceCarbonLakeriver the liqourix kernel is linux-zen kernel
I'm really digging the whole timeline splitting/naming. Lets me skip to what I really want to know.
@Richard Clutterbuck it is like a month but they didn't activate it for everyone at the same time
well same here is it new features??
I'm not seeing anything new on youtube android? Is it web or ios only?
@10:19 Exactly, frame times are more important than the highest fps.
Hi Chris. I think that it's essential in such tests to show lowest fps, not only highest and average.
Windows users: "I'm waiting for the update process to finish so I can use my computer again..."
Linux users: "Hey, let's install some of these cool custom kernels to see if I can get more fps in games!"
Let's face it, much of Linux is doing Linux for the sake of Linux.
@@TremereTT yeah. Its still has small userbase compared to the others so that's the first thing it needs. Once it gets big enough all kinds of support will automatically come.
Windows users: Oh I don't have a driver X, no problem windows is going to search and get it for me automatically.
Linux users: .... drivers ??? good luck reversing windows drivers :D :D :D
@@AbdelrahmanRashed more like Nvidia users lol
@@qwerasdfhjkio And printers in general.
Your GPU was pegged at 100% the whole time, so the only thing you're seeing here is any slight differences in the AMDGPU kernel driver... Try a CPU bound test to see real-world kernel differences...
this.
Your chosen benchmarking is GPU dependent, less effected by CPU.
If you use the same GPU drivers which are today generally distributed as kernel modules, of course you'll see nearly the same performance.
If anything,
This video is an example that you really have to know what you're doing to construct meaningful testing.
And, if you're going to test custom kernels or add or remove mods manually (Today's Kernel Loadable Modules can be loaded and unloaded by Users, it's only what is in the initrd you can't modify), you gotta know what's relevant or not to what you are doing.
Sorry Chris,
This was a good try but this video was likely educational in a very unintentional way.
...........................oh dear.......................
Thank you, Chris. I understand why people would enjoy hacking and futzing but my systems have always done what I want and more.
I am very impressed by a custom compiled Xanmod + BMQ scheduler (with some tweaked settings and more aggressive compilation flags).
Nice bootscreen, Shodan the Big Sister is watching us! XD
I use a first gen Ryzen 8 core and have been using the Liquorix kernel for stability. I find that I get fewer random reboots and system freezes with it compaired to the mainline kernel. I have not used, or heard of, Xanmod but I will be trying it out now.
Thx Chris, another good and informative video!
You forgot tkg-pds, which is only easily installed on arch, but is IMO the best custom kernel
I'd like to see tkg-pds as well
I don't know about that but I will try it out now. I don't mind compiling a kernel manually but I have to figure out how to make and use a RAM-disc.
the biggest difference being in the 0.1% minimums and general frametime consistency, something that has been pretty much ignored in this video
@@MGThePro Yes, people use FPS too much and framerendertime too little.
The best custom kernel is the one you configure specifically for your hardware.
i do use xanmod kernel and it makes a day and night difference for me. i love the project.
I was just thinking about custom kernels and to install one of them or not... 🤣 thanks Chris
I've been using the Zen kernel on Arch since you talked about it the other day. I play a lot of Rust and I can't say I've noticed any difference, but it works fine so I'll keep it anyway - they say it's better for general desktop usage as well as gaming.
I used too, apparently it focuses on keeping UIs running smoothly while the standard Linux kernel focuses more on services and processes because that's what you want on servers
@@dramm33 Me 3. I've never really noticed a difference and have both Zen and mainline on my machine. I've just installed Xanmod from the AUR so let's see how it goes.
Oh wow! There are sections in the timeline of the video!
I've recently switched to gentoo and the performance is great on my custom kernel already. Plus, the memory footprint of my kernel is as low as 85MB before I fired X server with DWM! I can play some windows games like blur on highest settings already but I need to explore more to see how gentoo performs in gaming.
I know this is an old comment, but you are probably more experienced now than when you posted this comment.
Did you switch from arch? Im considering switching because I like portage and compiling kernel on installation seems relatively straightforward. What was the biggest difference between Gentoo and your previous distro? And one last question, are you still on gentoo?
@@skeleskele I did switch from arch and the biggest difference I saw was the level of control one has over the machine which is incredible. Portage opens up a whole new level of possibilities due to being source based. As of now, I'm using Arch Labs since I have to complete a project fast and I want to use virtualization which I wasn't successful with my custom kernel. I'll go back to Gentoo because it's one of my favorite distros.
@@PixelTrik did you start from a defconfig or use a distribution .config? I HATE configuring the kernel myself. I'm a very indecisive person...id spend hours and hours in there enabling then disabling features then be like shit maybe I need that and re-enable them. I've taken the lazy route for the last 5yrs though...I use .config from archlinux then just change a few options that I know for a fact I need or don't need. Yes most ebuilds will check for kernel features it needs and will warn if it doesn't find them. Unfortunately not all ebuilds run that check. My kernel is probably heavier than it should be but configuring the kernel can be very brain racking. Its not really the "gentoo way" but its been working out ok for me. Yes we have distribution kernels in gentoo now too...but it's using a .config from fedora and alot seemed to be missing. And I've been a gentoo user for probably almost 15yrs.
@@jmills4313 I used defconfig to generate that.
The one time I was tempted to "roll my own" was for an oddball (secondhand) mobo built around a VIA Samuel CPU. That board gave me more problems! The CPU identified itself as 686 class, except that it in fact lacked CMOV and several other crucial instructions, which meant that livecd's with ISOLINUX compiled under generic 686 parameters would kernel panic while attempting to boot. As this was a majority of newly-released distros of the time, this forced me to check down to stripped-down 586-oriented (W2K styled) distros versus the 686-oriented XP-styled distros I preferred. Being fairly new to Linux back then, I lacked the sophistication needed to customize the ones that worked fully. In some ways, I still do. I eventually got more cooperative hardware, so the temptation passed. Said mobo went to a recycler. I'm still leery of VIA chipsets, as others have proven problematic under Linux as well. Has anyone else had this sort of experience with VIA?
Thanks for the video!! Because I work with pro audio, I was wondering how would Xanmod provide better latency results to record with lower buffer settings and with software monitoring
Thanks again
Great video...Btw... Have you tried Path of exile with the Vulkan Renderer ?
Thumbnail be like: "THE FOSS MUST FLOW"
Oh my god 😂 definitely this!
flowing foss
I personally use the TKG Kernel with the MuQSS scheduler. Even though it is only officially supported by Arch I have it compiled on openSUSE.
I was actually going to test xanmod patches on TKG (with real time patches) and see what I can come up with.
Gotta make the custom kernels even more custom c;
I ended up installing Xanmod for my ryzen laptop as a solution to a "CPU leak" I was having for KDE plasma.
turns out the kernel wasn't the problem, KDE had some messed up config or something that I found a command list that ended up fixing it properly.
but after installing xanmod I'll probably keep it just because it's a better kernel
I have been using mainline kernel s my whole life, but just recently I have found Liquorix. it gave me a noticeable speed increse while using desktop regularly, especialy resizing windows.
I would love to see more info on switching kernels at boot.
Also, does using a custom kernel hurt anything when doing other tasks?
Will Blender take a hit in performance, improve, or stay the same?
If it doesn't cost me anything, why not use a custom kernel all the time?
Switching kernels on boot is really easy. You just have to update GRUB for that. (Usually done by typing update-grub)
Since it's a different kernel some things may perform slightly different. Best would be for you to throw together some stuff you use often and test that.
The average FPS don't tell much as it's only GPU benchmark. No kernel was tested in your benchmark unfortunately...
For that to actualy be a kernel benchmark you must test response time speed (aka input lag) as these kernels aims to optimize latency. For that you must have some way to measure actual time of input and response on the screen. The faster, the better.
This is mainly because a lower fps but with low latency will generaly fells better to play than a high fps with higher latency.
Also, if you realy want to test how the kernel affects the performance of the CPU in games you'll also need to remove GPU bottlenecks. For a RX580 you need to run benckmarks at lower resolutions with every graphic detail set to the highest (without any AA) to enable every simulation the game uses. A 480p or even a 720p test should be enough to show that.
If you want to know more why playing games in Linux isn't good enough, here is a good study of the Linux Kernel latency that the Linux Scheduler causes: probablydance.com/2019/12/30/measuring-mutexes-spinlocks-and-how-bad-the-linux-scheduler-really-is/
He also compared to Windows results and that is actualy astounishing! Windows can respond almost 10 times faster than Linux and that's why games on Windows feels better. It's only a matter of latency.
For Linux to be good enough the main problem to be solved is latency. Then input lag.
Well, now that's the answer to "Is it ever too much blue?"
I just use the regular linux kernel since it has less compatibility problems.
Edit: I know it's not necessarily better but I use optimus manager since I have a laptop with hydro graphics (multi graphics). And it doesn't run even on zen kernel. I'm just not sure anything else would work.
I personally think Manjaro is the way to go for gaming.
Their default kernel is already tweaked (BFQ, AUFS, etc) for better gaming. As a bonus, it's basically Arch with some more testing for added stability: you can safely update Manjaro every time if you want. I prefer the KDE edition for a full fledged and light desktop enviroment, but the community editions are awesome and there is also an architect edition to build the system from scratch as you want it.
I'm not in any way affiliated with Manjaro, but i love it so much i have it installed on all my PCs, notebooks included.
Thanks for the video Chris. The Liquorix kernal does allow you to turn off hyperthreading and you may see a boost in cpu boost mode and faster speed on some games, but you can STS the hyperthreading off on any kernel and get the same result. Personally, I think you are splitting hairs.
I think there should be two kernel version one for laptop and another for desktop
but now its only server version
I just installed a bunch of custom kernels tonight, the Zen Kernel(same as liquorix i guess for Arch/Manjaro), Xanmod, and tkg-pds. All of them are quicker than stock. Zen wouldn't recognize my bluetooth devices though. In games i noticed the tkg-pds kernel was far superior then the other 2. Using the tkg-pds kernel for zen architecture(on my ryzen7 1700x) all my games ran buttery smooth Witcher 3 on ultra, Dragonquest 11, and Path Of Exile using the vulkan renderer added in the last patch. In fact these 3 games now run much better for me in Manjaro then they ran in windows 10! Linux gaming is really making insane leaps in performance!
Hey I love all of your videos.
I have one request will you please make a comparison video for Deepin Vs Linux mint.
Nice video! I think you may help me find how to force display through HDMI always on even if I unplugg and replugg the cable at booting or during the session.
Thanks to you, I'm a happy Manjaro user who plays window$ games. I play Armored Warfare, and I noticed that the longer I'm in the game, the more sound lags behind the video both in youtube and normal movies, and after a while it starts affecting the game too. I'm running the latest release candidate. I tried installing the 5.6 real time but that failed with a timeout. What would you suggest?
I'm learning all this programming stuff through osmosis. Regimented classes are my kryptonite. That being said, I feel really small right now. Hope I can eventually understand all this stuff.
What I know: Kernels work with compilers. Kernels are closely tied to the OS its being used on.
running linux(mainline), linux-pf (random 32bit software support), linux-fsync (futex-wait-multiple patchset for testing with Proton fsync). fsync works ok, but lacks some kernel mods for "normal" compiling. pf is just for random 32bit software support, ... i find and try strange software[[ don't judge me XD ]]. what did not go well (no details cus i'm not writing a blog) == linux-amd-staging-drm-next, linux-ck, linux-amd-git, and surprisingly (just to myself and my own opinions) linux-hardened wasn't as "hardened" as i was intending. Cool vid Chris. Rock out with your foss out. Peace.
Yay timestamps
Great to know! Now I will not even consider kernel changed and the basis of speed. Stability is where I'll be looking...
Are you testing in Virtual Machine or bare metal with harware screen recorder? How you are recording the dmesg info on shutdown?
I saw U do -- sudo reboot -- then the boot stopped to select a kernel. My boot displays a mostly black screen and does not stop. What is the secret sauce? U got me on Linux Mint Debian edition, away from Windows, with your series. Love your knowledge and teaching style. Newbie, me.
CTT, could you please elaborate on the lowest deviation xanmod. What it means and why?! For what purpose?! Thank you for the video.
hey bud, thanks for the awesome video but I'm facing a problem, after switching to liquorix kernel on Ubuntu18.04 hp laptop virtualbox module do not compile anymore which results virtualbox can't run without kernel, any suggestions?
This is actually something that I am working on right now. So far, I came to the conclusion, that there is no such thing as the perfect generic kernel. Let me explain: Xanmod is the best for my music production and for games. Having said that, the LQX kernel is not far off at all and if it is only about gaming then even the vanilla Arch kernel is looking pretty good. But when I compile stuff or do anything that just requires raw compute power then the best kernel so far seems to be the znver2 from the AUR. See where I am coming from?
All of the kernels have been compiled by me with rather aggressive custom CFlags in order to try to squeeze the last bit out of the machine. Next in line are the Xanmod-RT and the Xanmod-Cacule, the latter claiming that its CPU scheduler is superior to anything else.
But I already sense, that I will never find that one perfect kernel that fits it all. But maybe I can at least manage to reduce the amount of them because right now it kinda feels like if I had a dedicated kernel for each application.
I have a 4th generation intel with integrated graphics and pops I have thought of installing a custom kernel but I do not know if the low-end computers would get a performance improvement compared to the stock kernel that apparently sysem 76 got his hands on too
Chris I tried the Liquorix kernel with my ubuntu 20.04, and BOY did it just not work. It was almost comical the bugs I had. Fun to play with though. I was wondering if there is a custom kernel for Ubuntu 20.04 that would give me some better game performance.
I compile the kernel from source. You can customize it to your own hardware and performance needs. If you are prepared to work your way through all the options then you will out-perform any kernel provided by a third party.
You would probably enjoy Gentoo Linux... big time lol.
@@chesterhackenbush gentoo is a way of life. It becomes part of you.
Thanks for the video, but why one game? I am interested to know across a variety of titles as that way we can see whether this one game is an anomaly.
On Pop!_OS Liquorix had the lowest deviation and error. But I've noticed doing multiple tests I've gotten a lot of variation between the same Kernel. I even tried a benchmark with irqbalance off. But I'm still going to test them with actual gameplay and see if I feel a difference.
I'm using 5.4.43-1-MANJARO. As someone with aggressive ADHD, and using Linux to organize my madness, tweaking my computer is 1 of 20 different hobbies I'm constantly in and out of so it's pretty far down the list to play around with different kernels. Custom is a relative term anyways, this is Linux we're talking about.
if you use intel processor, you can add mitigations=off in grub or and cli 🙂
The name for this Kernel reminds me of a depressed elven mage in Baldur's Gate.
Life - is so hollow...
Nice shirt, I had a laugh there.
I would be curious about Mesa version and DXVK also
Not quite sure what you want to accomplish here, but ... doesn't it boil down to FPS vs INPUT LAG / controls responsiveness / accuracity? And not quite sure which one of them is actually for better game experience. The first one yes you can measure with some test suite, not the second one though. Yes, headless, controllerless server type kernel (mainline?) may with less CPU resources wasted to user input possibly renders the game faster (not in your example here, but theoretically) -- gives more FPS; but on the Liquorix kernel website even is the idea behind the kernel described as snappier and more responsive (a bit more resources "wasted" to actually better/faster input processing, that may or may not lose you some FPS, but overall improves the experience? Or find some kind of middle ground between mainline and real-time. So the real test here should be for INPUT LAG, and not necessarily only FPS ;)
Why change kernel if mainline Ubuntu is as good as the others? Have you tried with stock 5.4 kernel included in 20.04? Thanks
Custom kernels can perform better in most cases.
I for example had micro stutter on the vanilla mainline kernel while a custom kernel fixed it.
Good job. I would like to see as much of an apples to apples as possible. Minimize your time via run after hours over a minimum of 3 days before reporting the results. Script several different loads to exercise if an option. (but then again, maybe more trouble...) Anyway, just a passing thought. As Buffalo Springfield would say, "For What It's Worth" :~)
I think the same about custom kernels. I noticed games being smoother. Im using zen kernel.
Flightlessmango also did benchmark of some kernels, and there you can see the same, more stable while average fps may not be much better.
I have been reading some people say about how linux is stuttery in gaming and windows is better in that, i think part of that is due to default kernels not being good for gaming.
According to Phoronix test results, gaming is the is not where Xanmod and Liquorix really separate from mainline. All 3 gave relatively similar performance in gaming. Liquorix was slower than mainline everywhere else, and Xanmod was the fastest by a significant margin.
Atm I am just using linux-lts but I wondered how the linux-zen kernel matches up against it and also against those other custom kernels since it's in the official arch install guide
thought xenmod was my window manager :P xmonad
i use liqourix, I notice a difference in responsiveness but looks like the actual frame rate increase isnt that big
hey how is your timeline so coller on the video?
Would be interesting to throw XOrg vs Wayland into the mix...
Hey Chris.. I've an issue that my GPU is Nvidia GTX 1650 Ti 4GB, and I run Manjaro with KDE environment but I can't configure my graphic card. Have you a solution for this?
Does anybody compile their own kernels for gaming? If so what are the biggest factors? I'm thinking about messing around with preemptability and some other settings in Gentoo
I'm using Xanmod on my multi media server with Debian 10.
On my main machine i switched to Fedora just for distro hopping, and I dont miss the Debian based distros as much. But there is no good custom kernels for Fedora.
I dropped 6.7.10-1-liquorix-amd64 into Zorin 17.1 Pro, and as I'm on Intel graphics and pretty satisfied with that, I probably won't bother with XanMod. Still, cool video.
You didn't install the real time kernel for Xanmod. The RT Kernel is rt-edge. Even if you used the rt from xanmod I don't think one game is going to give anybody the results they need. But I think your doing some awesome videos for Linux new comers!
I notice very little difference between the experimental kernel and using zen, but I get way better performance and compatibility on laptop using the experimental kernel over the stable kernel in my Manjaro install.
Interesting, I have held off on running the experimental but can't know how it performs or acts if I never load it. Gonna see if the experimental fix a touchpad issue on laptop I have.
@@bigbay1159 Using the experimental kernel fixed an issue I was having with my keyboard. Might be worth a try.
Im on gentoo and because its a source based distro compiling your own kernel is part of the installation. I set the experimental use flag on and set keyword ~amd64 so i can pull in the latest stable kernel which is 5.18.14 at the time of this comment. Im using a ryzen 7 5700x and the experimental flag pulls in the specific cpu optimizations. Seeing that u can remove extra stuff in the kernel you dont need, i dont see a possible slight increase in performance being worth the stability of mainline kernel, especially if youre building it from source anyways.
I can`t use virtualbox after install xanmod kernel. Is this normal?
Is there anyone who has covered best kernels for laptops.... actual on battery most of the time usage scenario...
0:36 That shirt... LOL
How did you get chapters working in UA-cam? Absolutely brilliant feature!
I tried Liquorix and my WiFi stopped working. I tried installing drivers, etc., but it did not fix it. I reverted back to the mainline kernel and WiFi started working again.
I was just going to comment that often the tested and patched kernel for a distro gives the most stable and reliable performance. Tinkering can be fun, though. To each their own.
Forgot to test the Zen Kernel.. Personally, the difference is so small, it isn't worth the worry, nor the hassle... I stay on the Mainline if not the Lts...!
Using timeline will decrease your watch time. Am I right?
good test xD
Two things:
I don't consider Ubuntu default kernel mainline as it has some Ubuntu specific patches.
I don't think Xanmod, RT, -ck, zen are "custom" kernels as you do not customize them.
Custom kernel is the one you configure and compile yourself, maybe add some patches...
I laughed the 1st second seeing the text on your shirt
@UwU v2 Its a shell command to uninstall sudo with sudo-given privilege
wait you and I have effectively the same system? Wow, except I'm running the 5700XT over the 580 nowadays.
After running custom kernels on Ubuntu 20.0.4 my games run much smoother had some driver issues with resolutions on my first install all I had do was uninstall the driver and boot up the free driver than reinstall the non free driver and it worked fine.
Is it better to stick with the LTS kernel for stability if I'm on Arch?
I have it as backup. As in I have installed both the standard and the LTS so in the Grub menu I can pick the LTS if for some reason the standard gives me grief.
Could you show minimum frames or 10% lows next time?
You got it, this was my first benchmark using phoronix, so still learning the in's and outs.
Motoryzen but I think you tend to notice those lows more cause they tend to be so much lower. Even if the overall gaming experience follows the averages, the lows become noticeable.
Motoryzen I think 10% lows would last for more than a half a second
@@motoryzen a second is a pretty long time, actually. if you never notice things like stutter, or don't notice smoothness when there is no microstutter or "heartbeat", then i envy you? but an awful lot of people see over 30fps.. just sayin
WOW I'd crap my pants if shodan appeared like that on my grub screen!
"What are you trying to do _insect_... Do you even _think_ you can tinker with kernel-ker--kernels?"
Benchmark based on one game is not a benchmark
Lts-tkg-MuQSS all the way because I like to game while rendering. PDS has the best fps.
Is a custom kernel going to have a bigger effect on my performance when I'm CPU bound? My computer has a shitty CPU, and games would run much better if I could reduce the bottleneck caused by my CPU.
I switched to linux-zen for my manjaro, it gave me more stable fps in games via Proton
any updates mr titus on this subject ?
nice
edit: damn, now youtube supports chapter? that's cool
There are games that makes a huge diference like gta v i had a big jump like 15-20 fps but others like batman arkham knight i hardly see any difference but only on liquorix im having better performance, with xanmod i have the same performance as stock, im still trying understand this.
What about the zen kernel?
Hi, i really like your videos and u really helped me a lot before.
I was hoping you could help me resolve this issue, so kindly guide me.
I want to disable my F5 key on my laptop. I am running ubuntu 20.
i searched and mostly found this solution:
xmodmap -e "keycode 71 ="
or
xmodmap -e "keycode 71 = NoSymbol"
but this only disables the key inside terminal ...
in my chrome browser it still works.
Please help me solve this.
Thank you :)
There is no such thing as a mainline kernel. There are canned binary kernels packaged by distributions. If you want a real custom kernel you have to configure and build it yourself. Although how optimized you manage to make that is debatable. There is an almost incalculable number of permutations of the various options in config. The Linux kernel suffers from regression too. Later kernels do not often outperform earlier ones. It doesn't work like that.
Try to compile the kernel with the march=native patch set.
www.phoronix.com/scan.php?page=news_item&px=Linux-5.0-ad1-Optimize-CPU
This gives me a huge performance boost in bootup when all code is kernel. After reach userspace I didn't notice much.
For test you need systemd-analyze plot > plot.svg.
Liquorix ruined my audio in Linux Mint using an admittedly blah built-in audio setup. There was heavy reverb. The official kernel was fine as was Xanmod.
i wonder how this would be on a cpu heavy game like arma